Date: Wed, 10 Jan 2018 08:29:00 +0100
From: Willy Tarreau <w@1wt.eu>
To: Ingo Molnar <mingo@kernel.org>
Cc: linux-kernel@vger.kernel.org, x86@kernel.org,
        Andy Lutomirski <luto@kernel.org>,
        Borislav Petkov <bp@alien8.de>,
        Brian Gerst <brgerst@gmail.com>,
        Dave Hansen <dave.hansen@linux.intel.com>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        Peter Zijlstra <peterz@infradead.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Josh Poimboeuf <jpoimboe@redhat.com>,
        "H. Peter Anvin" <hpa@zytor.com>, Kees Cook <keescook@chromium.org>
Subject: Re: [RFC PATCH v2 3/6] x86/pti: add a per-cpu variable pti_disable
Message-ID: <20180110072900.GC14066@1wt.eu>
References: <1515502580-12261-1-git-send-email-w@1wt.eu>
 <1515502580-12261-4-git-send-email-w@1wt.eu>
 <20180110071951.t4vbwlnlw2qycnpm@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20180110071951.t4vbwlnlw2qycnpm@gmail.com>
User-Agent: Mutt/1.6.1 (2016-04-27)
Sender: linux-kernel-owner@vger.kernel.org

On Wed, Jan 10, 2018 at 08:19:51AM +0100, Ingo Molnar wrote:
> 
> * Willy Tarreau <w@1wt.eu> wrote:
> 
> > +#ifdef CONFIG_PAGE_TABLE_ISOLATION
> > +	this_cpu_write(pti_disable,
> > +		       next_p->mm && next_p->mm->context.pti_disable);
> > +#endif
> 
> Another pet peeve, please write:
> 
> > +	this_cpu_write(pti_disable, next_p->mm && next_p->mm->context.pti_disable);
> 
> or consider introducing an 'mm_next' local variable, set to next_p->mm, and use 
> that to shorten the sequence.

OK.

> More importantly, any strong reasons why the flag is logic-inverted? I.e. why not
> ::pti_enabled?

For me it's a matter of default case. Having a "pti_enabled" flag makes
one think the default is disabled and an action is required to turn it on.
With "pti_disabled", it becomes clearer that the default is enabled and an
action is required to turn it off. While it causes a double inversion for
the user due to the temporary choice of prctl name (we could have
ARCH_SET_PTI for example), I think it results on more readable code in
the sensitive parts like the asm one where these tests could possibly
end up inside #ifdefs. If we had "pit_enabled", something like this could
be confusing because it's not obvious whether this pti_enabled *enforces*
PTI or if its absence disables it :

#ifdef CONFIG_ALLOW_DISABLE_PTI
	cmpb $0, PER_CPU_VAR(pti_enabled)
	jz .Lend\@
#endif

But this is open to discussion of course.

Willy